Abstract: In location based social networks,friends prediction usually predicts friends with some similarity metric,and recommends the most similar users to some user．Traditional selection methods of user features don’t consider the difference between different features,so they cannot represent the overall features of users．This paper proposed a friends prediction method based on fusion of location information and social topology．First,we selected three relative features that can represent the overall user feature with information gain,then fused the selected relative features,and finally predicted friends with classification method．The experiments show that the proposed method doesn’t depend on the concrete classification method,and performs better than the multi-layer friend model.

Key words: Location,Topology,Recommendation,Link prediction